循环神经网络在结构损伤识别中的应用研究  

Application of Recurrent Neural Network in Structural Damage Identification

作　　者：王子凡 张健飞[1] WANG Zifan;ZHANG Jianfei(College of Mechanics and Materials,Hohai University,Nanjing 211100,China)

机构地区：[1]河海大学力学与材料学院,南京211100

出　　处：《河南科学》2021年第6期868-875,共8页Henan Science

基　　金：国家重点研发计划(2018YFC0406703)。

摘　　要：通过直接提取结构动态测试时序数据中的特征来实现结构的损伤识别,基于重力坝有限元模型在不同损伤情况下生成的加速度时序数据对循环神经网络进行训练和测试,以对循环神经网络在结构损伤识别中的应用进行研究.首先比较了传统循环神经网络(RNN)、长短期记忆(LSTM)循环神经网络和门控循环单元(GRU)神经网络的性能,然后采用网格搜索和随机搜索对LSTM循环神经网络的超参数进行优化.结果表明,相较于传统的循环神经网络(RNN)和门控循环单元(GRU)神经网络,LSTM循环神经网络在不同损伤工况下的识别准确率均有提升,最高达46.5%.数值试验结果表明在保证一定准确率的基础上,随机搜索比网格搜索用时缩短了85.2%,大大提升了搜索效率.The damage identification of the structure is realized by directly extracting the features in the structural dynamic test time series data.Based on the acceleration time series data generated by the finite element model of gravity dam under different damage conditions,the recurrent neural network is trained and tested to study the application of recurrent neural network in structural damage identification.Firstly,the performances of traditional recurrent neural network(RNN),long short-term memory(LSTM)recurrent neural network and gated recurrent unit(GRU)neural network are compared.Then,grid search and random search are used to optimize the hyper-parameters of LSTM recurrent neural network.The results show that compared with traditional recurrent neural network(RNN)and gated recurrent unit(GRU)neural network,the identification accuracy of LSTM recurrent neural network under different damage conditions is improved.The results of numerical experiments show that the time of random search is 85.2%shorter than that of grid search on the basis of ensuring a certain accuracy rate,which greatly improves the search efficiency.

关 键 词：循环神经网络 长短期记忆循环神经网络 损伤识别 网格搜索 随机搜索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TU312[自动化与计算机技术—计算机科学与技术]